Robust and Accurate Line- and/or Point-Based Pose Estimation without Manhattan Assumptions

Abstract

Abstract. Usual Structure from Motion techniques based on feature points have a hard time on scenes with little texture or presenting a single plane, as in indoor environments. Line segments are more robust features in this case. We propose a novel geometrical criterion for two-view pose estimation using lines, that does not assume a Manhattan world. We also define a parameterless (a contrario) RANSAC-like method to discard calibration outliers and provide more robust pose estimations, possibly using points as well when available. Last, we provide quantitative experimental data that illustrate failure cases of other methods and that show how our approach outperforms them, both in robustness and accuracy.

Region-Based Epipolar and Planar Geometry Estimation in Low─Textured Environments

“…Industrial images have a resolution of 1280×1920 pixels, whereas urban ones are 640×640. It must be noted that [9] fails to detect reliable segment correspondences in the considered examples. In the following, our method is compared to the standard method composed of two steps (i) extraction and matching of keypoints on the whole image by applying the Lowe's ratio on both directions then keeping only mutual matches (ii) computation of the epipolar geometry from this set of matched points using ORSA.…”

“…These four pairs of points thus define an homography induced by a vertical plane. Unlike [9], VPs are directly used here as primitives to define the models. That choice enables us to drastically reduce the combinatorics and the reliance on visual KPs, by limiting to just two the number of KP matches needed to generate an homography model.…”

“…Several types of matching strategies have been investigated in the past, and most of them consist in iteratively estimating a geometric model among a set of putative matches (RANSAC-like algorithms). Potential segment matches can be determined from the appearance similarity of their neighborhoods [7,8,9]. However, line segments in industrial environments often correspond to 3D object edges, i.e.…”

Multiscale line segment detector for robust and accurate SfM

“…Lines at window edges are also robust to reflection variations when the viewpoint changes, and calibration with lines naturally deals well with scenes that may be planar. As a matter of fact, a few line-based SfM methods [2], [9], [13] have recently been developed, that outperform classical pointbased methods on this kind of scenes, while not relying on segment endpoints, which are too uncertain with all known segment detectors.…”

“…II. RELATED WORK LSD [6] and EDLines [1] are currently state-of-the-art methods for line detection and are thus used in the mentioned SfM approaches [2], [9], [13]. They are based on an a contrario theory that automatically defines if a line is meaningful or not with a score called Number of False Alarms (NFA).…”